Disaggregating population density of the European Union with CORINE land cover

Abstract

This article describes and compares six disaggregation methods used to produce a dasymetric population density grid of the European Union at a 100 m resolution. Population data were initially available at commune level. The main ancillary information source was CORINE land cover, a land cover map distributed by the European Environment Agency. Information from the Eurostat point survey, land use/cover area frame survey, was also integrated in the parameter estimation of some of the approaches tested. Accurate population data for 1 km cell grids were provided by the Statistical Offices Sweden. These data provided the basic reference to quantify the accuracy of each method. The best results were obtained with a modified version of the limiting variable method (Eicher, C. and Brewer, C., 2001. Dasymetric mapping and areal interpolation: implementation and evaluation. Cartography and Geographic Information Science, 28, 125-138) that could be implemented, thanks to the national reference grids. For other methods the parameters could be estimated without using the reference grids; among them a method based on logit regression gave the best results. Compared with the traditional choropleth maps that represent a homogeneous density in each commune, the accuracy improvement of the disaggregated maps ranged between 20% and 67% (between 46% and 67% for the best method).

A high-resolution population grid map for Europe

“…To compensate for the fact that considerable amounts of urban and residential areas could be missing in the land use/cover maps due to generalization rules, cartographers (e.g. Eicher & Brewer, 2001;Gallego et al, 2011) found ways to assign population to vast cultivated and forested areas, thus creating less plausible population distribution patterns. To overcome this problem, an improved land use/cover dataset was developed and used to disaggregate population counts.…”

“…The first method (M1) is considered a 'control' approach because it replicates a known and documented disaggregation algorithm, as applied by Gallego et al (2011) and uses the original CLC as the only geographical ancillary data. On the other hand, two 'test' approaches (M2 and M3) make use of finer ancillary data combined with alternative disaggregation algorithms.…”

“…The Gridded Population of the World (Balk & Yetman, 2004) and the LandScan (Bhaduri, Bright, Coleman, & Dobson, 2002;Dobson et al, 2000) are two well-known global datasets that report population at respectively 2.5 and 0.5 minute-arc resolution. Efforts to produce a European population map have been conducted and reported by Gallego and Peedell (2001), Gallego (2010) and Gallego et al (2011). As reported in the cited papers, several algorithms were tested to disaggregate official European population counts (for the years 1990 and 2000) into 100 meter cells, using the Corine Land Cover (CLC) as main ancillary dataset.…”

“…choropleth mapping (see, for e.g. Eicher & Brewer, 2001;Gallego, Batista, Rocha, & Mubareka, 2011;Holt, Lo, & Hodler, 2004;Langford, 2007;Mennis & Hultgren, 2006). In addition, dasymetric mapping solves the distortions caused by the Modifiable Areal Unit Problem (Openshaw, 1984).…”

“…An incomplete list would comprehend the following: binary allocation (Langford, Maguire, & Unwin, 1991); limiting variable estimation (Eicher & Brewer, 2001;Gallego et al, 2011); empirical estimation of local densities (Mennis, 2003;Mennis & Hultgren, 2006); regressions (Briggs, Gulliver, Fecht, & Vienneau, 2007;Flowerdew & Green, 1989); kernel functions applied to point data (Bracken & Martin, 1995;Martin, 1989); estimation-expectation algorithm (Dempster, Laird, & Rubin, 1977, and its applications such as in Flowerdew, Green, & Kehris, 1991;Gallego et al, 2011;Harvey, 2002) and geostatistical estimations (Kyriakidis, 2004;Liu, Kyriakidis, & Goodchild, 2008;Wu & Murray, 2005). The paper of Wu, Qiu, and Wang (2005) systematizes and reviews most of existing population estimation methods.…”